Medical health care is undergoing a significant change of paradigm. Moving health care from health centers to home environments poses new challenges for acquisition, management and mobile exchange of information. The HealthNet project at RWTH Aachen University has developed a prototype which addresses these new challenges: a Body Sensor Network (BSN) collects information about the vital functions of a patient while she is in her home environment; the integration of smart textile sensors increases the acceptability of such technology; mobile communication and data management enables the exchange of health data between patients and doctors; data stream mining techniques tuned for mobile devices provide immediate feedback of the collected data to the user; and finally, advanced security and privacy features increase user acceptance and cope with legal requirements. This paper summarizes the challenges and achievements in the development of this prototype.
The demographic change with a growing population of
elderly people and the associated increase of health care
related expenses require new models for health care
management. Moreover, there is a growing group of health-aware
people that would like to take more personal responsibility
for their own health, e.g., by monitoring their vital
parameters during sport activities. New innovative technologies are
necessary to fulfill these new requirements. Mobile and
remote health monitoring has demonstrated positive influence
on patients disease courses, especially for chronic diseases
[
In this paper, we describe an extendable and flexible
monitoring system for the case study of physiological state
monitoring of runners. The system has been developed in the
context of the HealthNet project [
Using a mobile communication infrastructure (e.g., UMTS, LTE, or Wi-Fi), mobile devices can communicate with each other such that peers can easily exchange health information. Especially, the mobility of patients is improved as detailed monitoring can now be performed at home: periodically or in the case of important events, the device sends the collected and pre-processed data to information systems maintaining patient health records (e.g., hospital information systems) which can be accessed by medical experts.
The main challenges in this project are • the design and development of wireless medical sensors, which are able to monitor vital functions of a person, • the integration of theses sensors into textiles and development of electronic units as textiles (e.g., conductive paths) for unobtrusive and comfortable usage, • the integration of the data collected by various sensors in one data stream, and • the analysis, mining, and aggregation of the sensor data to detect emergency events, to reduce communication costs, and to predict near future.
We addressed these challenges in the HealthNet project and report in this paper our experiences in developing an integrated prototype. Sectioin 2 first describes the requirements analysis which we have done with a group of athletes. An overview of the prototype system and its architecture is given in Section 3. The main components of the prototype are an intelligent T-Shirt with integrated conductive leads/electrodes (cf. Section 4) and a Body Sensor Network (IPANEMA, Integrated Posture and Activity NEtwork by Medit Aachen) which aggregates multiple data streams from a range of sensors (cf. Section 5). The data are transmitted wirelessly over a Bluetooth interface to a mobile device for visualization and a first lightweight analysis (cf. Section 6). A more detailed analysis of the data is done on a server which receives periodically or in case of peculiar events data from the mobile device. We also performed a case study in a running event of which we will briefly summarize the results in Section 8.
For gathering of requirements, four active runners on semiprofessional level were interviewed. All interviews were conducted by two interviewers with one interviewee. The interviews used a unique set of 14 questions regarding mobile health monitoring, and eight questions regarding a stationary counterpart. The questions especially targeted the usability of smart phones as supporting device in runs and the information needs of the runners and the willingness to share information during training and competitions. All interviews took about one hour, and were recorded for postinterview analysis. 2.1
The interviewed runners are male, three between 20 and 25 years old, and one between 30 and 35 years old. All runners participate in competitions on national level. The disciplines range from 3000 meters steeplechase to marathon distance, and triathlon. All interviewed persons do intensive training between 5 and 15 sessions per week. 2.2
2.2.1
All interviewed persons consider their self-assessment as the most important source of information, which is even more reliable than any physiological measure. They stated to ignore measured high peaks of their pulse if feeling good, and also low measures if feeling bad. That means, that their subjective rating of their state is more important for them than an objective measurement. They treat many technologies as fun, which could be more interesting to increase motivation in mass sports. 2.2.2
In addition to the self-assessment, all interviewed runners were interested in heart rate (they all measure it in training). The data could be used to trigger a notification if exceeding upper or falling below lower personal limits. Furthermore, the sportspersons consider breathing rate and oxygen absorption relevant to detect exhaustion in advance. Sweat analysis could lead to an estimation of water balance of the body, useful for reminding the runner to drink or to determine the amount of liquid to drink for convalescence after a training session. Determining the blood sugar level could signal a low sugar level or hunger knock2. Last but not least, all interviewed runners had measured lactate in the past. It is probably the best indicator for the current personal fitness. Noticeably, none of the interviewed runners associated any value with blood pressure information, even if explicitly asked by the interviewers.
Most measurements listed above require settings incompatible with daily outside use. Some request tests in medical labs, some include in addition blood analysis (such as blood sugar level, lactate) which is not compatible with mobile use. All interviewed runners agreed that therefore it will be challenging or impossible to apply these measurements in their training sessions or competition. 2.2.3
Speed measurements have a strong influence on the running speed. All runners pointed out that speed measures from cars or bicycles are not usable because of the meaningless unit (mph, km/h) and low precision. They request measures of time needed for the last lap (on cycle tracks), the last 400 meters or the last 1000 meters (all preferable in a unit of minutes:seconds) to adjust their personal running speed accordingly. 2a completely run out of energy, also known as “bonk” or “hit the wall”
For uphill sections, the absolute distance and remaining distance of the uphill part are valuable for all interviewed runners. The gradient is less important because of the low absolute number.
Other information, like weather conditions, weather forecast or condition of the ground are important in preparation of training or competition; it is of no value while being on the move. 2.2.4
All interviewed persons request methods for personalisation of the measurements and accompanied items, such as frequency, upper/lower border. 2.3
2.3.1
All interviewed runners had applied technology for monitoring heart rate; all interviewed runners knew technology for gathering track data (i.e., GPS). None used other technology, like step counters or sensors in shoes. Only one person carries the mobile phone in training sessions, in a back pocket together with keys. They do their sports without listening to music.
All interviewed runners track heart rate in training, only two do the same in competitions. Two do not track the heart rate in competition mainly because of loosing comfort, i.e., chest belt slipping out of place and making the runner feeling confined. The interviewed runners do not agree to carry any additional device. In competition, none of them would be willing to carry a mobile device. 2.3.2
Carrying a mobile device while doing sports is considered burdensome. There must be a reasonable benefit from doing so. It must not require any attention by the runner, it must not swing (e.g., on a neck strap), it must not disturb the rhythm of arms, legs or breathing (the latter nearly excludes speech interfaces). The device must be lightweight, small, waterproof and shock resistant. The touch-sensitive surface, if any, must come in a sweat resistant cover.
The shape and feeling of a watch was considered most appropriate, as applied in current monitoring systems for heart rate. It can be worn at one arm and operated with the hand of the other arm. If more functions are to be integrated, the only sensible way of carrying a larger mobile device seems to be a pocket at the arm. It supports a similar way to operate it using the hand of the arm not carrying the device. 2.3.3
Operation of a 1-button-watch was considered sufficient; nevertheless the operation of buttons of a mobile device were considered to require too much attention and too fine granular movements for hand and finger. A mobile device at the arm can be similarly operated by touch on the display.
The interviewed persons see the problem with touching the display that it might get dirty and smeared by the runner’s sweat, making checking current values from the display impossible. Because of disruption of rhythmic breathing, speech-based operation is only considered feasible for a few short commands. 2.4
2.4.1
Live-sharing information with others is considered a minor issue by the interviewed runners. Together with personal trainers, post-processing (for long distance runs) and frequent analysis after smaller sessions (e.g., in interval training) was seen to be more important than live data transmission. One interviewee had the idea that the trainer might interrupt over-pacing of a runner in a hopeless intermediate state of a competition, especially if it is one in a row of competitions. All interviewed persons declined to lively share personal or medical information with other external persons like friends, training mates, online communities, or event organisers or competitors. It was only acceptable for notification in case of emergency.
To receive information from others, trainers and supporters call out time information and intermediate state of the competition to the runner on track. The interviewed runners think that receiving more information, e.g., about personal state of competitors, is rather distracting. One of the interviewed persons stated a value of knowing intermediate state of competition within the same age group, in particular if persons nearby are of the same or another group like the runner. Getting the positions of the team mates was considered not interesting, neither in training nor in competition.
As an open question, the interviewees brainstormed about other ideas for valuable live-sharing of information. As a result, it could be valuable for optimisation of the handover in relays, especially in long distance relays. It would be of value to the successor to know the personal health state of the predecessor in order to adjust warming and preparation phase. If the predecessor is in good shape, the estimated arrival time is earlier than if the person is in bad shape, influencing the point of time to start preparation. 2.4.2
After a training session or competition, the runners were open to share track and time data with team mates and online communities, which is already implemented by portals like http://www.gpsies.com. 2.5
Post-processing of the collected data is very important to all interviewed runners. They asked to file all information to a computer system for persistent storage. They all use a kind of training diary, two use already computer applications for this purpose. 2.5.1
The interviewed runners asked for easy connection with the PC, and easy to handle download. 2.5.2
All interviewed runners do intensive performance analysis combining tracking data, time data, health information, and comments on personal feeling. If applicable they compare current data with past datasets for recurring events, competitions, tracks, or distances. The main goal of the analysis is identification of flaws in performance (absolute speed, endurance, power to go uphill) requesting updates of the training method and plan.
The triathlete analyses shifts in performance of the single disciplines, e.g., intensively training one discipline has contradictory influence on the performance in the other two.
One runner mentioned to use the post-processing also to estimate lifespan of used hardware, e.g., professional running shoes that loose suspension after 3000 km of use, demanding for replacement to prevent damage from tendons and ligaments. 2.6
Based on the requirements analysis, several use cases were identified which are described in this section. The use cases are grouped in four categories: Sensor management, mobile monitoring, sharing, and archiving. 2.6.1
The sensor managing use cases describe the setup, configuration and maintenance of the set of sensors delivering information to the system. The actor usually is the user. In addition, other persons or organizations might perform the use cases as well, e.g., an emergency doctor who adds a sensor after the user had an accident, or a physician who adjusts the upper border of a physiological parameter to raise notification earlier. The actor employs a plug-in / plug-off mechanism to add or remove sensors to the network; this should be as automatic as possible. The added sensors perform registration and de-registration at the controlling component of the sensor network. Configuration of the sensors should be also possible, so that user can adjust the properties of the sensor (e.g., sampling rate, sensor identifier, measure unit, data transmission rate) to his/her personal needs. 2.6.2
The monitoring use cases describe the use of a mobile system to monitor the health status. The user employs the system for observing specific parameters, being informed about the current status and alarming himself or another entity during a personal activity. The user can also turn off all monitoring and notification functions by muting the device. 2.6.3
The group of sharing use cases describes the information exchange between all parts of the system with external entities (e.g., server or other users). It applies to sharing information while being mobile as well as sharing information from the other parts of the system like the archive. The group contains: 2.6.4
The archiving use cases describe the use of and retrieval from a persistent storage. The user employs a stationary device (such as a laptop or desktop PC) to search for information of a specific type, date and time, activity, or value. The archiving use-cases are:
The HealthNet prototype is based on a BSN integrated into a textile platform (i.e., T-shirt) measuring the physiological state of a person. An overview of the system is illustrated in Fig. 1. A registry server manages the communication between different peers in the network. The sensor data is received by a smartphone via Bluetooth which sends the data to other peers in the network. Other peers in the network are an advanced data mining & analysis service or other trusted parties such as trainers and doctors.
In the current prototype, the BSN consists of an ECG sensor, a combined temperature/humidity sensor, two 3D acceleration sensors, and a master node. The master node collects the data from the individual sensors and sends it to the smartphone. Conductive yarn acts as electrodes as well as leads. The signals are received by the ECG sensor attached to the shirt. The sensor processes the ECG and infers the current heart rate from it.
On the smartphone, a mobile application integrates the health data with data measured by the phone, such as the current GPS position. The mobile application also visualizes, stores, and analyzes the data. If enabled by the user the integrated data is sent via UMTS or Wi-Fi (IEEE 802.11) to a registry server which distributes it to registered third parties, such as a trainer, a doctor, or a server analyzing the data. The architecture also allows sending feedback and results of the analysis of the data to the users smartphone. 4.
The state-of-the-art electrodes used for most medical applications are, for example, disposable electrodes glued onto the skin. These electrodes are coated with electrolyte-gel to improve the conductivity. The advantages of those electrodes are low contact impedance and a fixed position. However, they are not suitable for a continuous long-term measurement because the electrolyte-gel can dry and may also cause allergic reactions. Moreover, the wires between electrodes and the sensor exacerbate the handling for untrained users. To achieve the aim of a continuous and mobile monitoring system, another solution has to be found.
Textile electrodes could be a good alternative for the standard ones. They can be used for long-term measurements because they are not coated with electrolyte-gel. The yarn for the textile electrodes must possess high conductivity, good elastic behavior to assure a good skin conformance and it should be biocompatible due to the constant skin contact. Another advantage of textile electrodes is that these electrodes can be integrated into garments which lead to a very high mobility of the whole system and intuitive handling. Mobility can be further increased by using textile integrated conductive paths instead of cables. A reversible interface is necessary to remove the sensor node before washing. However, textile electrodes also have disadvantages: the contact impedance is higher and movement causes motion artifacts.
Suitable yarns matching all requirements mentioned above have been researched and tested. The best one was a silvercoated polyamide yarn. A circular foam padded textile electrode with a radius of 2.5 cm was used. In addition to the ECG electrodes, the same material was also used to manufacture the textile conductors (see Fig. 3). The textile conductors were applied to the outside of the T-shirt with metal push buttons to connect both electrodes and the sensor. Preliminary results with this T-Shirt show the suitability of textile electrodes for the application as ECG electrodes.
Body Sensor Networks (BSN) usually consist of a varying number and diverse types of sensors. They are wirelessly connected either to each other, called mesh network, or to a central master node, called star network. The acquired data is then transferred over wide area networks (WAN) to Master node Sensor
Light‐weighted Single‐stream
Mining & Analysis Smartphone
Registry Server central data and health service providers for further processing. This section focuses on the challenges in developing the medical sensors, connecting them in a BSN, and processing the measured signals.
Bringing health status monitoring to personal health care environments presents a new set of challenges: devices have to be small, unobtrusive and easy to handle. Preferably, they need no or only minor interaction and are connected via wireless technology to the supervising medical professional or health care center.
The IPANEMA BSN is designed to be easily modified for
different application scenarios, e.g., cardio-vascular
monitoring or hydration status monitoring [
The sensors of the current prototype produce a raw data stream of about 14 kbit/s which is transmitted over a 433 MHz ISM band transceiver with a proprietary protocol.
The network is structured in a star topology. The leaves are formed by a flexible number of modules which can be equipped with different types of sensors. The sensor data is send over-the-air to a central master module. The main tasks of the master node include network management, data transfer to a mobile device and creating time synchronization beacons for the sensor nodes.
The goal in the design of the architecture of the mobile application was to have a very flexible and extensible system. As explained before, the HealthNet project is not limited to a particular application domain, our solution should be applicable in a healthcare domain as well as in a sports domain. To allow easy customization and adaptation to new domains, we identified four main components for the mobile application on the smartphone (cf. Fig. 4).
The HealthNet Controller is the central unit for managing the set of active sensors, and notifying dependants if measures changed value or the composition of the network changed. The Data Cache stores recent sensor data in a Graphical User
Interface
Configuration Single-Stream
Prediction Data Cache
Data Window
Data Transmission
Unit Preprocessing for Multiple Streams
Prediction
Incoming/ Outgoing Data
Measure Sensor
HealthNet
Controller Data Window such that a single-stream prediction over a short timeframe is possible. The windows are implemented as a circular data structure - if a window is full, the latest incoming data will flush the oldest data. Furthermore, the cache stores also all data (if desired by the user) such that the data can be uploaded to a server for detailed data mining and analysis later on.
The Data Transmission Unit (DTU) takes care of the information exchange among different stakeholders. Four methods of sending data to authorized entities have been implemented. Any external entity must prove eligibility to receive any data from the mobile application. The DTU supports three communication modes: 1. Request-response: an external entity requests information from the mobile application. The DTU retrieves the requested data from the cache and transmits the response. This is for example done when a trainer wants to see detailed data about a runner. 2. Time-based submission: A fixed interval after which a selected data set is sent, e.g., data is sent from the runner to a trainer only every 10 seconds to reduce required bandwidth and communication costs. 3. Direct transmission: The relevant data is transmitted directly to the receiver. This mode is used for audio feedback from a trainer to a runner.
Due to the modular design, peer mobile applications use roughly the same architecture, with the only difference that these applications receive data via the DTU and not from sensors.
On the user interface level, the data which is received from the sensors or other peers is managed according to the use cases as described in section 2. 6.1
To get the maximum benefit of the HealthNet application the measured data has to be analyzed to detect critical situation or events, and to make a short term predictions. Data mining techniques in this context are restricted by two important constraints: (i) the data is a continuous stream and has to be analyzed in real-time; persistent storage and long time series of data are not available as in classical data mining tasks, (ii) the resources (CPU power, battery life, memory) of the mobile device are very limited.
To cope with the problem of limited resources, we
developed: (i) an adaptive technique for anytime classification,
which is capable of both, classifying under varying time and
resource constraints, and incrementally learning from data
streams to adapt to possible evolutions of the underlying
data stream [
Context prediction is an emerging topic in the field of
data mining, e.g., predicting the location of mobile objects
was a frequently tackled subtask of mobile context
prediction in recent researches. For scenarios of managing health
information of mobile persons, the prediction of the near
future health status of persons is at least equally important
to predicting their location. A first method for predicting
the next health context of mobile persons equipped with
body sensors and a mobile device has been developed and
implemented [
Our main observation is: frequent sequential patterns appearing in rules containing multiple streams, are completely built using frequent patterns existing in each single stream. Thus, predicted values were directly presented to the user in the mobile application using a light-weighted resourceaware algorithm that was implemented locally on the user’s mobile phone. More accurate predicted values were sent to the user from a multiple stream prediction algorithm which was implemented on a server using the preprocessed frequent patterns on each stream (cf. Fig. 4). 6.2
A rigorous evaluation of security and privacy risks was
done, requirements were derived from it, and the
implementation was developed accordingly [
Confidentiality during data collection is achieved by using ZigBee AES-128 encryption between the sensor nodes and the master node, and Bluetooth encryption E0 is used between the master node and the smartphone. Confidentiality during data storage is achieved using AES-128 encryption on the devices, so that no data can be recovered wrongfully by someone who has physical access to a device. During communication between trusted devices, we do not rely on the security mechanisms of the technologies used (e.g., UMTS, LTE, WLAN) because the data must not be revealed to the network operators, and wireless technologies such as UMTS, LTE, and WLAN typically only encrypt the air interface. Instead, all data transfers apply AES-128 encryption and message authentication codes on the application layer.
In wireless connections to trusted parties, all parties are identified using certificates with shared keys. The implementation of the encryption is transparent to the application as standard interfaces of the Android SDK are used to implement secure storage and communication. In addition, we found that the authentication and encryption mechanisms had no significant influence on battery life or performance of the handheld device. 7.
The interest in mobile healthcare applications started with
systems like [
A reasonable idea to integrate real-time monitoring into
daily life activities are the application of wearable or textile
sensor platforms. This section therefore reviews the
integration of sensors into garments, such as sport shirts or similar.
In [
To integrate multiple vital parameters into one textile platform, several sensors are combined to form a sensor network. Often, a master component controls the network and centralizes data acquisition, short-term storage and transmission. These can be realized either wired or wireless. 7.2
The MyHeart-project3 led by Philips was dedicated to the
prevention, diagnosis and therapy of cardiovascular diseases
The monitoring is based on sensors integrated into daily
life textiles, such as undergarment. A sensor shirt has been
3http://www.hitech-projects.com/euprojects/
myheart/
developed using conductive and piezoresistive yarn for
monitoring of heart (ECG) and respiratory activity (impedance
pneumography), core and skin temperature with non-textile
sensors and an accelerometer [
The respiratory sensing technology was also used in the
Wealthy project4 [
A project that supports medical treatment and behaviour
of elderly people suffering from cardio-vascular disease is
described in [
Commercial products are available on the market in particular to support ambitious sports(wo)men. The products do not aim on sophisticated measuring medical data. Usually, it is considered sufficient to provide heart rate and calories burned, and location and time related information. The often use wrist or chest bands.
A large set of wrist-mounted computers is available for example from Polar, ranging from low-end technology for beginners to high-end systems for professionals like the RS8005. They receive body signals from chest straps, display and store the information on the watch, and allow for downloading and post-processing with the personal computer. A similar system is the Garmin Forerunner6. It monitors time, distance, pace, heart rate and calories burned. As Garmin’s unique selling proposition, it additionally tracks the position of the sportsperson by the use of a high-sensitive GPS receiver built into the wrist watch. The GPS antenna is partially integrated into the watchstrap. The heart rate is measured by the use of a chest strap. The system supports different profiles, e.g. for swimming, cycling, and running of triathlons.
A more sophisticated system is the adidas miCoach7. It is an integrated system to plan, work-out, and analyse personal training. As the main part of the system, it combines three components to support the work-out: an auditive display (miCoach Pacer) for heart rate measures, speed and distance which reacts to the speed; a bundle of a chest belt measuring heart rate and a wristwatch (miCoach Zone); an application running on the user’s mobile phone for coaching (miCoach Mobile).
As a main advantage, the textile strap for monitoring the heart rate can be replaced by two different bra’s (adidas supernova glide/sequence bra) or a shirt (adidas supernova cardio shirt). Nike+8 is a training system similar to miCoach developed by Nike and Apple. The main differences to miCoach is that Nike+ combines the features from Nike’s 4http://www.wealthy-ist.com 5http://www.polar.fi/en/products/maximize_ performance/running_multisport/RS800CX 6http://www8.garmin.com/marathon/forerunner/ 7http://www.micoach.samsungmobile.com/ 8http://nikerunning.nike.com/ running shoes with an integrated step counter (with the drawback of getting depended on the Nike’s brand), and that it uses the iPod instead of a mobile phone (with the same drawback of dependency). 8.
We implemented an end-to-end prototype for a runner scenario (training and competition mode) with one or more runners and a trainer. Case studies with the implemented prototype have been conducted during the Lousberglauf 2011 & 2012 (a local running event in Aachen with about 2000 participants). A team of five runners has been equipped with the sensors and smartphones. In addition, a trainer monitored the performance of the runners using also a smartphone. Data communication and management did not cause any problems; the trainer could always see the position and vital parameters of the runners. Due to excessive motion artifacts during running, we used standard electrodes for the run. In the meanwhile, we did some additional measurements with a new version of the textile electrodes in a lab environment on a treadmill which gave better results. We also improved the algorithm for inferring the heart rate from the raw ECG data, such that it is less sensitive to movement artifacts. This improved the data quality in the second case study in 2012, but the data quality is still too low for deriving health-related advices.
We have shown in this project that health monitoring using mobile wearable sensor networks is feasible. Data management and analysis can be done in real-time although the data is coming at a high frequency. Security and privacy issues have been addressed by implementing suitable encryption and authentication mechanisms into the application. In another related project (Nanoelectronics for Mobile AAL-Systems9), a similar approach for data management has been developed in the context of Ambient Assisted Living (AAL). Some results (e.g., the architecture of the mobile application in Fig. 4) have been applied also in this project.
However, we have seen that with the current technology, problems like data management, analysis, security, and privacy can be solved as mobile devices are powerful enough in terms of CPU and communication bandwidth. The real challenges are at the two ends of the data processing flow: firstly, the sensor data must have very high quality to be useful in any kind of application (for sportspeople or patients), false alarms will be annoying, missed alarms might be fatal; secondly, the potential users have to be convinced about the usefulness of such technology. In our interviews, the sportspeople were sceptic about the benefit of such an application. The same applies also to elderly people who might be even more reluctant in wearing any device that monitors them.
This work was supported by the DFG Research Cluster of Excellence on Ultra High-Speed Mobile Information and Communication UMIC (http://www.umic.rwth-aachen.de) at RWTH Aachen University.